Botulinum toxin type A for the prevention of headaches in adults with chronic migraine

Original submission

3.10 The manufacturer carried out a systematic review for economic evaluations of botulinum toxin type A for people with chronic migraine, but no relevant published studies were identified.

3.11 The manufacturer developed a Markov model using the pooled data from the two PREEMPT trials, comparing botulinum toxin type A with placebo. Only approximately 35% of the population met the criteria set out in the NICE scope (people whose condition has failed to respond to three or more prior pharmacological prophylaxis therapies). Because of the similarity in efficacy, (see section 3.9) the manufacturer used the population whose condition had failed to respond to one or more pharmacological therapies, which was approximately 64% of the trial population, in order to increase the statistical power.

3.12 Six health states were defined in the model according to the number of headache days per month: 0–3 (health state 1), 4–9 (health state 2), 10–14 (health state 3), 15–19 (health state 4), 20–23 (health state 5) and 24 or more (health state 6). Patients could enter the model in any of the three chronic migraine states (health states 4,5 or 6); chronic migraine was defined as 15 headache days or more per 28 days. The cycle length was 12 weeks, which is the same as the frequency of administration of botulinum toxin type A in the trials. The model used the transition probabilities observed in the clinical trials, which include the transitions between the six health states as well as an additional transition to discontinuation of therapy. For the placebo arm, the second cycle transition probability was repeatedly applied after week 24. For the botulinum toxin type A arm, the same transitional probability matrix was applied for the third, fourth and fifth cycles; this figure was calculated by averaging these three cycles. For people continuing on treatment in cycles 6–9, the same average figure was used. The time horizon used in the model was 2 years, in line with the manufacturer's expected maximum treatment duration with botulinum toxin type A. For the first of the 2 years, trial data were used if possible, with extrapolation used for the second year, using the last observations from the trial.

3.13 As well as using the discontinuation data from the clinical trials in the model, two treatment (dis)continuation rules were applied. A negative stopping rule was applied at week 24 for people whose condition responded insufficiently to botulinum toxin type A. Insufficient response was defined as people having an improvement of less than two health states (a reduction of between 5 and 10 headache days per 28 days depending on both the number of headache days and the corresponding health state at the start of treatment) after two successive cycles of treatment. After discontinuation, they moved between health states using the transition probabilities for placebo treatment, but did not incur treatment costs. A positive stopping rule was applied at week 48 for people whose condition had changed to episodic migraine (fewer than 15 headache days per 28 days). The model assumed that after 1 year, people whose condition had changed to episodic migraine stopped treatment with botulinum toxin type A and remained stable in the episodic migraine health states thereafter.

3.14 Two mapping equations were derived to map the MSQ data (as collected during the trials) to EQ-5D utility, so that utilities could be attached to each health state. One equation was used to predict utilities for people in the chronic migraine health states, and the other was used for predicting utilities in the episodic health states. Utility was different for each health state in this model and also between treatments within the same health state, the latter being justified on the grounds that treatment with botulinum toxin type A was shown to affect a broad range of relevant outcomes such as the severity and intensity of headaches, as well as the number of headache days. Utility was predicted for each health state using the patient-level data from the trials. This was done by grouping patients by health state and calculating the mean utility per treatment group. Utility values used in the model ranged between 0.479 and 0.746 depending on the treatment and on the number of headaches experienced. Adverse events affecting health-related quality of life were captured in the model by applying discontinuation rates. There were assumed to be no additional costs relating to adverse events.

3.15 The manufacturer assumed that chronic migraine in the relevant patient population would be managed in a specialist setting, with associated consultant-related costs. Administration time was assumed to be 30 minutes, at a cost of £73.00. Cost of treatment was calculated based on one 200 unit vial of botulinum toxin type A at £276.40, with no vial sharing, leading to a total cost of £349.40 per 12 week cycle. For people in the placebo arm (standard care), the cost was assumed to be £36.50 per 12 weeks, which is based on a 15 minute appointment with a consultant every 24 weeks to optimise acute therapy. Costs for GP visits, accident and emergency attendance, hospitalisation and the cost of triptan were modelled for each health state but only three different costs were used; for health state 1, states 2–3 and states 4–6. These costs came from a poster by Bloudek et al. (2011) of Scottish resource use according to the International Burden of Migraine Study (IBMS). Costs and quality-adjusted life years (QALYs) were discounted at a rate of 3.5% per year.

3.16 For people whose condition failed to respond to at least three prior preventive medications, the total cost and QALYs for placebo were £1895 and 1.20 respectively, compared with a total cost of £2438 and 1.29 QALYs for botulinum toxin type A. This gave an incremental cost and incremental QALYs of £543 and 0.09 respectively and an ICER of £6083 per QALY gained for botulinum toxin type A compared with placebo. There was a 69.1% probability of botulinum toxin type A being cost effective if the maximum acceptable ICER was £20,000 per QALY gained. The manufacturer did not report a central estimate of the ICER calculated by probabilistic modelling. For the wider population, which included people whose condition had not responded to one or more prior preventive medications, the costs and QALYs were similar, with an ICER of £5828 per QALY gained for botulinum toxin type A compared with placebo.

3.17 The manufacturer presented a variety of sensitivity analyses, and for most of the cases the ICER for botulinum toxin type A compared with placebo stayed under £10,000 per QALY gained. Excluding the positive and negative stopping rules increased the ICER to £15,294 per QALY gained. Restricting the time horizon to 1 year increased the ICER to £14,098 per QALY gained.

3.18 The manufacturer identified an error in its analyses, in which the original ICERs were based on utilities derived from all people in the trials. It submitted revised ICERs using utilities from people whose condition failed to respond to at least three prior preventive medications. After correcting for the error, the ICER for botulinum toxin type A compared with placebo increased from £6083 to £6434 per QALY gained.

ERG critique of original submission

3.19 The ERG noted that the trials were generally of good quality. The ERG pointed out that in the PREEMPT 1 trial, at baseline patients in the botulinum toxin type A group had a statistically significantly lower frequency of migraine episodes (11.5 compared with 12.7, p=0.006) and frequency of headache episodes (12.3 compared with 13.4, p=0.023), but had statistically significantly more cumulative hours of headache on headache days (295.7 compared with 274.9, p=0.022) than those in the placebo group. The original primary outcome in PREEMPT 1 was to have been frequency of headache episodes, but this was changed to headache days because of new guidelines for the conduct of clinical trials in chronic migraine. This was considered reasonable by the ERG.

3.20 The ERG found botulinum toxin type A to be effective regardless of the number of previous preventive medications taken. As the number of previous preventive medications rose, so did the relative effectiveness of botulinum toxin type A compared with placebo (reductions of 2.3 and 2.7 headache days per month for the one prior preventive medication and three or more prior preventive medications subgroups respectively). In addition, the ERG observed that the placebo effect decreased as the number of previous preventive medications increased.

3.21 The ERG stated that a noticeable feature of the trial data was the size of the improvement on placebo, which lasted for at least 24 weeks and was only modestly less than on active treatment. In addition the ERG pointed out that the efficacy of botulinum toxin type A was greatest with the first injection and that the second injection had much less effect, with similar efficacy to that of the second placebo injection. The ERG therefore queried whether a negative stopping rule could be applied after the first injection.

3.22 The ERG's main concern about the design of the PREEMPT trials was about whether blinding was maintained and if not, what the effect of this was on the results of the trials. The ERG pointed out that in previous botulinum toxin type A trials, 70% of patients receiving botulinum toxin type A correctly guessed what they had received, because of changes in muscle tone such as reduced wrinkling of the forehead. The ERG further explained that because unblinding is an important factor in controlled trials of preventive treatment of chronic migraines, the International Headache Society guidelines recommend that subjects and investigators should be questioned at the end of trials about whether they thought active or placebo was administered during the trial. This was not done in the PREEMPT trials.

3.23 The ERG considered the manufacturer's model to be reasonable for the decision problem. The ERG noted that the model and trial data sets for headache days per month at week 24 corresponded poorly and that this overestimated the net impact of botulinum toxin type A by 53%. Yet this overestimate had no direct impact on cost effectiveness because it was the distribution between health states that determined the costs and QALYs within the model. The ERG observed that within the validation data, the model and trial were reported by the manufacturer as having no patients in health states 5 and 6 at week 24 in the botulinum toxin type A arm. The ERG noted that this could be because of the negative stopping rule being applied in both sets of data, with these patients falling into the discontinuation category, but could not verify this. Uncertainty remained around this point and also the distribution between health states among those patients modelled as discontinuing because of the negative stopping rule, but who were followed up within the trial. The ERG noted that there were no clinical data to support the use of the two stopping rules. The ERG queried whether the negative stopping rule at 24 weeks in the model would apply in practice, and whether the requirement for an improvement could be based on either one or two health states. It found that an improvement of only one health state within two cycles increased the ICER for the three or more prior preventive medications subgroup from £6083 per QALY gained to £8354 per QALY gained. The ERG doubted that the positive stopping rule at week 48 was likely to be implemented. It queried how long people who had stopped treatment would remain stable before needing re-treatment, noting 2-year follow-up data from a published abstract by Rothrock et al. (2011), which estimated the duration of response to treatment with botulinum toxin type A based on a 50% reduction in the number of headache days per month. Of a total sample of 100 people whose chronic migraine responded to treatment with botulinum toxin type A, 68% maintained a good response to treatment with continued injections at a frequency of 3.2 months. A further 24% were able to stop treatment and maintain a good response for at least 6 months; and 8% relapsed back to chronic migraine. The ERG found that removing the positive stopping rule roughly doubled the ICER for the three or more prior preventive medications subgroup for botulinum toxin type A arm compared with placebo, from £6083 per QALY to £12,542 per QALY gained.

3.24 The ERG noted that there were wide disparities between the model inputs for resource use used in the economic model, which were based on the poster by Bloudek et al., and those reported in a published paper by Blumenfeld et al. (2011). The ERG noted that the two sources had reported results from the same international study, however Bloudek et al. appeared to include costs based on Scottish implementation, with no information on the number and ages of the patients, GP visits, hospital outpatient and accident and emergency attendances, hospital admissions or medication use. The ERG noted that because the higher costs for health states 4–6 were driven by hospital admission costs and were the same in each of the health states 4–6, details of the Bloudek et al. poster were important because it could be that the manufacturer's estimate of resource use was too high. It was for this reason that the ERG preferred the resource use detailed in Blumenfeld et al. which contained more detailed data. The ERG had concerns about the applicability of international data on hospital admission rates for chronic migraine to clinical practice in England and Wales, but recognised that Blumenfeld et al. contained the most detailed data available.

3.25 The ERG observed that utility values were calculated using MSQ estimations at 24 weeks, when the full effect of botulinum toxin type A had occurred. The ERG noted that the manufacturer's model generally assigned higher utility values for the botulinum toxin type A arm than the placebo arm for each health state and that this feature of the model accounted for about half of the anticipated QALY gain for botulinum toxin type A. It noted that for this to be valid, there would have to be a difference in other features such as: the number of headaches experienced during a headache day, the duration of these headaches, or the proportion of headaches that were migraines. The ERG noted that this was plausible, although it was unsure whether the substantial difference in utilities in each health state was justified given the small absolute differences in secondary endpoints in the PREEMPT trials. The ERG commented that the number of headache days per month was not included in the estimate of utilities, which instead relied on IBMS MSQ data. The manufacturer had outlined a regression analysis from the EQ-5D in the IBMS and found that the EQ-5D and the number of headache days per month were statistically significantly negatively correlated. The ERG used these data and showed very little difference in utilities between botulinum toxin type A and placebo in a given health state. The ERG queried whether the IBMS MSQ data could be used to estimate utilities given that the IBMS included a different population to that described in the decision problem. It noted that a more appropriate estimation of utilities would encompass both MSQ and HIT-6 data. The ERG also explained that there is non-monotonicity (an anomaly in which the utility values are not consistently increasing or decreasing) in the utility values in the botulinum toxin type A arm in which people in health state 5 (20–23 headache days per month) have a lower utility than those in health state 6 (24–28 headache days per month). The ERG commented that this may highlight some uncertainty between the mapping of MSQ data to utilities and correspondence with the model structure as defined by the number of headache days per month.

3.26 The ERG noted that the model included the cost of administering botulinum toxin type A and of placebo, based on 30 and 15 minutes of neurologist time respectively. The ERG stated that review of a patient and administration of botulinum toxin type A may take between 30 minutes and 1 hour, depending on the experience of the person injecting and the possibility of complications during administration. As a result the ERG commented that the cost of administering botulinum toxin type A based on 30 minutes of neurologist time may be too optimistic. It also noted that the appropriate cost of follow-up outpatient consultations should be the standard NHS neurological outpatient follow-up reference cost of £140. The ERG noted that administration of botulinum toxin type A was reported to be offset by cost saving because of reductions in GP visits, inpatient admissions and accident and emergency attendance, but it considered the amount of the cost saving to be overestimated.

3.27 The ERG explained that the model submitted by the manufacturer was non-linear, with a central probabilistic estimate of cost effectiveness more than double that of the deterministic estimate, based on ERG exploratory analyses. It stated that some elements of the probabilistic modelling seemed unwarranted; for example, treating the identity matrix as having an uninformed prior and modelling it probabilistically. The ERG found that removing these elements reduced the degree of non-linearity but did not eliminate it. The ICER for the three or more prior preventive medications subgroup increased from the manufacturer's deterministic ICER of £6083 per QALY gained to the probabilistic ICER of £11,447 per QALY gained if the ERG's revisions were applied. If the ERG's revisions were not applied the probabilistic ICER was estimated at £14,004 per QALY gained.

3.28 The ERG conducted a number of different exploratory sensitivity analyses. In these analyses, the ERG used the manufacturer's model with a number of modifications. The modifications included:

3.29 The ERG reported that applying all the above modifications (using the manufacturer's utility values based on one prior preventive treatment) resulted in an increase in the ICERs for botulinum toxin type A compared with placebo to give a deterministic ICER of £10,257 per QALY gained and a probabilistic ICER of £16,165 per QALY gained. The ERG reported that using these assumptions there was a 67% probability of botulinum toxin type A being cost effective if the maximum acceptable ICER was £20,000 per QALY gained, and an 87% probability of botulinum toxin type A being cost effective if the maximum acceptable ICER was £30,000 per QALY gained.

3.30 The ERG also explored the impact on the deterministic ICER of varying the negative and positive stopping rules and applying the same utility values for each arm. In these analyses, all the modifications outlined in section 3.28 were applied and the utilities from the three or more prior preventive medications subgroup were used. The ERG also used 2-year follow-up data from the published abstract by Rothrock et al. (2011) (see section 3.23).

3.31 Applying the same utilities from the three or more prior preventive medications patient subgroup to both arms (based on the updated utility data from the manufacturer) resulted in lower QALYs for botulinum toxin type A with no change in costs, compared with the base case. When the manufacturer's base case with positive and negative stopping rules and different utilities were used, the deterministic ICER for botulinum toxin type A compared with placebo was £11,267 per QALY gained, compared with £20,324 per QALY gained when the same utilities were applied to both arms for each health state. When the positive stopping rule was excluded (compared with using the positive stopping rule for people who move from chronic to episodic migraine) the ICER for botulinum toxin type A compared with placebo increased from £11,267 to £19,483 per QALY gained when different utilities were used in the botulinum toxin type A and placebo arms. When a positive stopping rule was applied to 24% of people who had moved from a health state of chronic migraine to episodic migraine (based on the study by Rothrock et al.) the ICER for botulinum toxin type A compared with placebo increased from £11,267 to £17,438 per QALY gained when different utilities were used. The ERG also applied various negative stopping rules: no negative stopping rule, an improvement of at least one health state within the first cycle; improvement of at least one health state within two cycles; and the manufacturer's base case of an improvement of two health states in two cycles. When no negative stopping rule was applied and different utilities were used in the botulinum toxin type A and placebo arm, the ICER for botulinum toxin type A compared with placebo was £13,986 per QALY gained. When different negative stopping rules were applied, the ICER for botulinum toxin type A compared with placebo ranged from £9618 to £13,167 per QALY gained. The scenario that had the most impact on the model results for botulinum toxin type A compared with placebo was when no positive and negative stopping rules were applied. This led to an ICER of £24,387 per QALY gained if different utilities for the three or more prior preventive medications patient subgroup were applied to both arms, or £46,290 per QALY gained if the same utilities for the three or more prior preventive medications patient subgroup were applied to both arms. The ERG also remarked that applying separate estimates for the transition probabilities for cycles 3, 4 and 5 to be more in line with patient data would further increase the ICERs by 5–10%.

Additional submission after consultation

3.32 The manufacturer provided additional analyses after consultation in response to NICE's request for a revised economic model based on the Committee's preferred assumptions, and some clarification of a number of issues relating to the calculation of utilities in the model, the large difference between the deterministic and probabilistic ICER, and resource use. The manufacturer presented:

In addition the manufacturer provided analyses varying the time horizon and placebo effectiveness, and also the results of an Australian survey of 10 people who received treatment with botulinum toxin type A for their chronic migraine.

3.33 The revised manufacturer's model presented the cumulative deterministic results for each of the Committee's preferred assumptions on the manufacturer's original deterministic ICER of £6083 per QALY gained when different or the same utility values were applied for each arm:

3.34 The manufacturer's revised cumulative results were based on a negative stopping rule of a two health state reduction in two cycles of treatment, as assumed in their original model. The manufacturer conducted one-way deterministic sensitivity analyses to assess the impact of different negative stopping rules on the ICER (from no negative stopping rule to a stopping rule of less than 50% reduction in headache days per month increasing in 5% or 10% increments from 0%). The ICER varied between £14,198 and £17,174 per QALY gained when different utilities were applied, and from £23,849 and £30,755 per QALY gained when the same utilities were applied. The ICER when no stopping rule was assumed was £19,508 per QALY gained when applying different utilities, and £35,637 per QALY gained when applying the same utilities. The manufacturer considered a negative stopping rule of a 30% response rate based on the reduction in the number of headache days per 28 days after two cycles of treatment to represent its new base case, based on evidence that it is used in clinical practice in the UK for the treatment of chronic migraine. When using the revised base-case assumptions and inputs (see section 3.33) and a 30% negative stopping rule, the revised base-case deterministic ICER was £14,999 per QALY gained when different utilities were applied, and £24,939 per QALY gained when the same utilities were applied.

3.35 The manufacturer explored the causes of the large difference between the deterministic and probabilistic results observed in the manufacturer's original submission and the ERG's original additional analyses. The manufacturer identified one reason for the difference to be the method used to sample the transition probabilities for the probabilistic sensitivity analysis. In their original submission, these were calculated using a Dirichlet based sampling. When the manufacturer recalculated the probabilistic values using a beta tree sampling method, the central probabilistic estimate and the deterministic estimate were broadly in line with one another. The manufacturer also identified an error in its application of a continuity correction. When using the revised base-case assumptions and inputs, a 30% negative stopping rule, beta tree sampling, and correcting for the continuity correction error, the probabilistic ICER was £14,959 per QALY gained (compared with a deterministic ICER of £14,999 per QALY gained) when different utilities were applied, and £25,242 per QALY gained (compared with a deterministic ICER of £24,939 per QALY gained), when the same utilities were applied.

3.36 In response to NICE's request after the first Committee meeting, the manufacturer provided additional clarification about the differences in resource use between Bloudek et al. and Blumenfeld et al. The manufacturer highlighted that the primary reason for the differences was the inclusion of data on Brazilian patients in Bloudek et al. The manufacturer also provided one-way deterministic sensitivity analysis using data from the UK subset of the IBMS. When using the revised base-case assumptions, inputs including a 30% negative stopping rule, and resource use data from the UK subset of the IBMS, the deterministic ICER reduced from £14,999 to £12,624 per QALY gained when different utilities were applied, and from £24,939 to £20,989 per QALY gained when the same utilities were applied. The manufacturer stated that the reduction in the ICER was caused mainly by the higher than average number of hospitalisations in the UK patients in the IBMS.

3.37 The manufacturer carried out additional sensitivity analyses that varied the time horizon and the effectiveness of placebo, assuming no retreatment. The cost effectiveness of botulinum toxin type A was sensitive to the time horizon, with the deterministic ICER ranging from £14,999 to £8825 per QALY gained when the time horizon ranged from 2 to 20 years, and when different utilities were applied. When the same utilities were applied, the deterministic ICER ranged from £24,939 to £11,700 per QALY gained when the time horizon ranged from 2 to 20 years. The manufacturer also performed a sensitivity analysis in which it was assumed that people in the placebo arm remained in their original health state throughout the trial. This was done because the placebo effect observed in the trials would not be expected to exist in clinical practice in people with chronic migraine who have tried three or more prior preventive headache medications. When no effectiveness in the placebo arm was assumed, the ICER reduced from £14,999 to £5677 per QALY gained when different utilities were applied, and from £24,939 to £6008 per QALY gained when the same utilities were applied to each arm.

3.38 In response to NICE's request for clarification on utilities, the manufacturer provided further details of the mapping algorithms used in its economic model, which were applied to the week 24 data of the PREEMPT trials in the three prior preventive medications group. The manufacturer also supplied mean values of the coefficients and of the parameter values of the week 24 data. The manufacturer stated that the rationale for applying different utility values to treatment arms comes from the results of the PREEMPT trials, which showed significant clinical and health-related quality of life benefits beyond the reduction in the number of headache days per 28 days.

3.39 The manufacturer provided results of an Australian survey of ten people with chronic migraine who had received botulinum toxin type A for up to 5 years. The survey indicated that treatment with botulinum toxin type A provides three main types of clinical benefit: reduced frequency of migraine and headache days; reduced severity of migraine; and reduced length or duration of migraine symptoms. Participants also identified additional benefits to treatment, such as reduced reliance on other treatment, less sick leave, fewer accident and emergency and GP visits, as well as overall improvements in personal, social and mental well-being.

ERG critique of the additional submission

3.40 The ERG reviewed the manufacturer's consultation comments and revised economic model. The ERG noted that it might be more appropriate to assume a placebo administration cost of £70 based on 30 minutes of neurologist time compared with a cost of £140 based on 60 minutes of neurologist time, given that the latter cost and time would involve the administration of botulinum toxin type A as well as a consultation. The ERG also justified a lower cost for neurologist time for placebo administration, in view of the fact that patients whose chronic migraine has failed three prior prophylaxis therapies have regular follow-up in some but not all neurology clinics. The ERG's analyses showed that when a placebo administration cost of £70 was applied to the manufacturer's revised base case (including a 30% negative stopping rule) the deterministic ICER increased from £14,999 to £19,244 per QALY gained when different utilities were applied. No ICER was calculated for applying the same utilities to each arm.

3.41 The ERG noted that the utility mapping functions were constructed to pick up elements of the condition that are not captured by the frequency of headache days per 28 days (such as severity or intensity of headaches). Therefore given that health states are defined in terms of the number of headache days per 28 days, it was not surprising that there were non-monotonic utility values. The ERG explained that there are alternative methods for calculating utilities to those chosen by the manufacturer, such as restricting the individual MSQ dimensions to being monotonic or at least non-decreasing, which could reduce or avoid the non-monotonicity observed in the original utility values. It noted that the manufacturer's utility mapping functions have different non-MSQ parameter values between health states for botulinum toxin type A and placebo. The ERG stated that the manufacturer had not provided a rationale as to why these parameter values should be different. When the ERG equalised the non-MSQ parameter values in the mapping functions, less non-monotonicity was observed and the ICER was £18,895 per QALY gained when different utilities were applied to each health state. The ERG stated that when the individual MSQ dimensions were restricted to being monotonic or at least non-decreasing, the ICER ranged from £17,278 to £18,256 per QALY gained.

3.42 The ERG reviewed the manufacturer's scenario analysis in which people in the placebo arm remained in their original health state throughout the model. The ERG argued that the benefits of botulinum toxin type A include a large placebo effect, therefore it would be wrong to retain the placebo effect when receiving botulinum toxin type A but not when receiving the placebo. The ERG also reviewed the scenario analysis that varied the time horizon. It noted that no clinical trial data existed beyond 12 months, as observed in the PREEMPT trial, but a 2 year time horizon is plausible. It also reviewed the manufacturer's Australian survey of people (n=10) who received botulinum toxin type A for their chronic migraine. The ERG expressed concern that the study included a very small number of participants who were paid to take part, and who received botulinum toxin type A for up to 5 years, which indicates that a positive stopping rule was not applied. It noted that the survey participants reported a good response to treatment with a dose of 100 to 150 units of botulinum toxin type A.

3.43 The ERG reviewed the manufacturer's comments about resource use. The ERG expressed continuing concern that it is unclear how to interpret some of the resource use data from Blumenfeld et al. and Bloudek et al. When reviewing the resource use data for people from the UK with chronic migraine in the IBMS presented by the manufacturer, the ERG noted that the subset included a small number of people (n=57), and a significant number of outlying results (for patients whose resource use contributed to most of the total resource use observed). The ERG stated that the manufacturer should have explored the impact of excluding outlying results from its analyses, and that given the small number of people from the UK with chronic migraine in the IBMS, they considered the Blumenfeld et al. resource use data to be most applicable to the UK.

3.44 The ERG identified an update of the published abstract by Rothrock et al. (see section 3.23). The update, published by Hanlon et al. (2011), closely matched the results published by Rothrock et al., with 25% (compared with 24% in Rothrock et al.) of people able to stop treatment and maintain a good response for at least 6 months after stopping treatment. It noted that 24–25% was the most appropriate figure on which to base a positive stopping rule.

3.45 Full details of all the evidence are in the manufacturer's submission and the ERG report, which are available from www.nice.org.uk/guidance/TA260